CoolFace
Modelpublic

lianghsun/Llama-3.2-Taiwan-3B-Instruct

sourceHugging Facellama3.2updated 5mo agoView on Hugging Face
28likes45downloads
Model Card

Model Card for lianghsun/Llama-3.2-Taiwan-3B-Instruct

<!-- Provide a quick summary of what the model is/does. --> <a href="https://discord.gg/fj6WbHMvfs" target="_blank">[👋 歡迎加入 Discord 討論 🎉]</a>

image/png 圖像生成來自 [OpenArt](https://openart.ai/home)：An anime-style 🦙 standing proudly atop the summit of Taiwan’s [Yushan (Jade Mountain)](https://zh.wikipedia.org/wiki/%E7%8E%89%E5%B1%B1), gazing forward.

採用 lianghsun/Llama-3.2-Taiwan-3B 為基礎模型（foundation model），使用大量中華民國台灣的繁體中文對話集和多國語言對話集進行模型指令微調（instruction fine-tuning）和多輪迭代直接偏好優化（direct preference optimization, DPO），旨在訓練出具有中華民國台灣知識及風格的小語言模型（small langugae model, SLM）之對話模型。

<details> <summary><b>Model Change Log</b></summary>

Update DateModel VersionKey Changes
2025/01/01v2025.01.01Fine-tuning is based on the foundation model version v2024.12.28, and it uses self-prepared instruction datasets for this round of fine-tuning.
2024/12/13v2024.12.13Completed 1st round DPO training (10/10 epochs). Preparing for next round DPO training.
2024/11/27v2024.11.27Completed SFT training (5/5 epochs). Preparing for multi-round DPO training.
2024/11/25v2024.11.25Updated model version to v2024.11.25, training progressed to (3/5) epochs. Still in SFT stage, DPO training remains pending.
2024/11/22v2024.11.22Initial upload: Model version v2024.11.22, training completed up to (1/5) epochs. Currently trained only on SFT, DPO training not yet performed.

</details>

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

Model Sources

<!-- Provide the basic links for the model. -->

yaml
user: 請介紹台灣
assistant: 台灣，位於亞洲東部，地處太平洋與菲律賓海之間，面積約36,000平方公里，人口約2,300萬，是民主自由的國家，經濟實力強勁，擁有世界第10大經濟體。台灣以美食、文化、自然美景著稱，還有豐富的歷史與傳統，吸引全球遊客。台灣語為官方語言，但中文也廣為使用，英語也常用於國際交流。台灣政治多元，執政黨為民進黨，台灣是全球科技產業的重鎮，擁有先進的製造業與服務業。台灣氣候溫暖潮濕，四季分明，夏季炎熱，冬季涼爽，雨季則在5月至10月。台灣的美食以小吃為主，如滷肉飯、珍珠

Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. --> 本模型已經具備有繁體中文對話能力，使用者可以直接部署推論端點使用。

Downstream Use

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app --> 若需強化模型在特定領域的知識，可透過微調進一步提升其性能與專業能力。

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. --> 本模型旨在提供資訊，不參與任何政治或法律問題的評斷或立場表達。

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. --> 語言模型的生成內容可能因訓練集的多樣性而帶有偏見、特定立場，或包含與事實不符的言論，請使用者務必在使用過程中仔細確認內容的準確性與中立性。

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

要使用 vLLM Docker image 來啟動此模型，您可以按照以下操作：

bash
docker run --runtime nvidia --gpus all \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HUGGING_FACE_HUB_TOKEN=<secret>" \
    -p 8000:8000 \
    --ipc=host \
    vllm/vllm-openai:latest \
    --model lianghsun/Llama-3.2-Taiwan-3B-Instruct

請注意，如果想要使用不同版本的 checkpoint，請加上 --revision <tag_name>

bash
docker run --runtime nvidia --gpus all \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HUGGING_FACE_HUB_TOKEN=<secret>" \
    -p 8000:8000 \
    --ipc=host \
    vllm/vllm-openai:latest \
    --model lianghsun/Llama-3.2-Taiwan-3B-Instruct --revision <tag_name>

Training Details

Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

<details> <summary><b>繁體中文對話資料集</b></summary>

</details>

<details> <summary><b>多國語系對話資料集</b></summary>

</details>

Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

Preprocessing

(WIP)

Training Hyperparameters

<details> <summary><b>SFT stage for v2024.11.27</b></summary>

Note: 以下包含 v2024.11.22 和 v2025.11.25 的超參數設定

  • —learning_rate: 5e-05
  • —min_learning_rate: 5e-07
  • —train_batch_size: 105
  • —seed: 42
  • —distributed_type: multi-GPU
  • —num_devices: 4
  • —gradient_accumulation_steps: 50
  • —total_train_batch_size: 21,000
  • —optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • —lr_scheduler_type: cosine
  • —lr_scheduler_warmup_ratio: 0.01
  • —num_epochs: 5.0
  • —global_step: 590 </details>
Speeds, Sizes, Times

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. --> <details> <summary><b>SFT stage for v2024.11.27</b></summary>

Note: 以下包含 v2024.11.22 和 v2025.11.25 的超參數設定

  • —Duration: 5 days, 16:15:11.17
  • —Train runtime: 490,511.1789
  • —Train samples per second: 25.37
  • —Train steps per second: 0.001
  • —Total training FLOPs: 26,658,386,120,540,160
  • —Train loss: 0.8533 </details>

Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

Testing Data, Factors & Metrics

<details> <summary><b>Formosa Taiwan Knowledge Bench</b></summary>

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->

lianghsun/Formosa-bench

#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

[More Information Needed]

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

[More Information Needed]

### Results

[More Information Needed]

#### Summary

</details>

<details> <summary><b>lianghsun/tw-legal-benchmark-v1</b></summary>

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->

#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

[More Information Needed]

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. --> Accuracy

### Results

  • —Model Revision: v2024.11.27
**Subset****Split****Score**
lianghsun/tw-legal-benchmark-v1train31.1

#### Summary

</details>

<details> <summary><b>tmmlu++</b></summary>

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->

  • —Dataset: ikala/tmmluplus
  • —Revision: c0e8ae955997300d5dbf0e382bf0ba5115f85e8c

#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. --> [More Information Needed]

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. --> Accuracy

### Results

  • —Model Revision: v2024.11.27
**Subset****Split****Score**
geography_of_taiwantest36.2
dentistrytest33.83
technicaltest35.07
statistics_and_machine_learningtest28.57
clinical_psychologytest29.6
tve_designtest38.54
three_principles_of_peopletest48.2
introduction_to_lawtest29.96
linear_algebratest21.43
agriculturetest24.5
jce_humanitiestest38.89
musictest25.9
secondary_physicstest33.04
physicstest27.84
advance_chemistrytest27.64
junior_science_examtest30.05
veterinary_pathologytest25.09
financial_analysistest25.13
national_protectiontest42.65
macroeconomicstest26.76
politic_sciencetest27.44
ttqav2test61.06
junior_chinese_examtest30.86
traditional_chinese_medicine_clinical_medicinetest25.9
junior_math_examtest21.71
auditingtest21.82
anti_money_launderingtest37.31
pharmacologytest30.68
trust_practicetest28.18
tve_mathematicstest18.67
human_behaviortest32.04
pharmacytest22.76
tve_chinese_languagetest36.65
optometrytest25.11
physical_educationtest30.73
organic_chemistrytest35.78
tve_natural_sciencestest33.73
educationtest37.9
mechanicaltest42.37
taiwanese_hokkientest14.73
nautical_sciencetest30.49
business_managementtest39.57
logic_reasoningtest27.34
marketing_managementtest39.78
economicstest25.95
basic_medical_sciencetest28.41
occupational_therapy_for_psychological_disorderstest35.73
general_principles_of_lawtest31.13
junior_chemistrytest24.88
veterinary_pharmacologytest36.3
educational_psychologytest33.52
finance_bankingtest32.59
official_document_managementtest32.43
fire_sciencetest30.65
junior_social_studiestest47.62
accountingtest20.94
engineering_mathtest27.18
education_(profession_level)_test.csv)test24.07
chinese_language_and_literaturetest27.64
management_accountingtest24.19
culinary_skillstest39.38
administrative_lawtest25.71
insurance_studiestest33.42
real_estatetest22.83
computer_sciencetest31.61
taxationtest27.47
tradetest20.32

#### Summary 模型版號 v2024.11.27，無論是基礎模型（lianghsun/Llama-3.2-Taiwan-3B）還是指令微調模型（lianghsun/Llama-3.2-Taiwan-3B-Instruct），均未接受過 tmmlu++ 資料集的訓練，以確保測試的公平性。經測試，目前該模型在 tmmlu++ 上表現普遍不佳，未達及格分數，可能需要加入專業領域的資料集來強化基礎模型能力。

</details>

Model Examination [optional]

<!-- Relevant interpretability work for the model goes here -->

[More Information Needed]

Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Technical Specifications

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware
  • —CPU count: 32
  • —Logical CPU count: 64
  • —GPU count: 4
  • —GPU type: NVIDIA H100 NVL
Software
  • —OS version: Linux-5.15.0-124-generic-x86_64-with-glibc2.35
  • —Python version: 3.12.7

Citation

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

bibtex
@misc{lianghsun2024llama32taiwan3binstruct,
  author = {Huang, Liang Hsun},
  title = {Llama-3.2-Taiwan-3B-Instruct},
  year = {2024},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/lianghsun/Llama-3.2-Taiwan-3B-Instruct}},
  note = {Accessed: 2024-11-25}
}

Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. --> N/A

More Information

Acknowledge

image/png 在此致謝鴻鵠國際股份有限公司蔡長明先生無償地贊助算力，以及曾經幫忙過：廖振翔、chweng、Ben、kevin、Maxxchu、Lam 和陳林彥…等朋友們，才能讓這個模型得以訓練完成，提供算力者乃人生父母。

Usage

如果你基於此指令模型進行微調，希望能不吝嗇在 模型卡片(model card) 裡標註 基礎模型 為：

yaml
base_model: lianghsun/Llama-3.2-Taiwan-3B-Instruct

標註和 ❤️ 是給予我們最大的鼓勵，謝謝。😀

Model Card Authors

Huang Liang Hsun

Model Card Contact

Huang Liang Hsun

Framework versions

  • —Transformers 4.45.2
  • —Pytorch 2.4.1+cu121
  • —Datasets 2.21.0
  • —Tokenizers 0.20.0